AI has created the intelligent driving "veteran driver"

In the process of changing the real world, autonomous driving has become one of the widely discussed scenarios, and its development direction to some extent reflects whether the era of true general artificial intelligence can accelerate.

At the Yunqi Conference on September 19th, He Xiaopeng, Chairman of XPeng Motors, and Wu Xinzhou, Global Vice President and Head of the Automotive Business at NVIDIA, held a roundtable forum to discuss the disruption of AI to the automotive industry.

He Xiaopeng pointed out that end-to-end autonomous driving is the first large-scale implementation of the entire big model. Internet giants such as Google and Baidu have entered the field one after another, but have not successfully promoted the large-scale implementation of intelligent driving until the emergence of end-to-end.

He believes that previous intelligent driving relied on human-written rules to operate, but "a person cannot face all the scenarios in the world with rules. End-to-end big models can raise the ceiling of future intelligent driving higher and improve the lower limit." He stated that from now to the next 36 months, intelligent driving after getting on board with end-to-end can make car owners drive like experienced drivers, which users can deeply perceive.

Wu Xinzhou agreed with this, further explaining that the understanding of the physical world after training the big model far exceeds that of humans. "With such universal capabilities, end-to-end can handle various complex scenarios."

It is worth mentioning that the popularization of end-to-end in the industry began with Tesla's introduction of the FSD V12 version last year. Now, the process of FSD entering China is already underway, and whether domestic players as "followers" will be disrupted has become a common issue that the industry needs to face.

In response, Wu Xinzhou stated, "The biggest role of Tesla's FSD is to drive industry-wide leapfrog development." The problem is that FSD has not completely solved the lower limit issue, the original logic errors still exist in FSD, and there is still a lot of work to be done in the face of China's complex traffic conditions.

He Xiaopeng expressed his expectation for FSD entering the Chinese market. He pointed out that after FSD uses end-to-end big models, the industry has seen a huge change, with a more human-like, smooth experienced driver, smoother and more efficient than driving on his own. "We need different technologies to impact this market, so that everyone can see the change."

He Xiaopeng asserted that AI integration will overturn the pattern of the automotive industry, "there will be fewer players at the table in the future. In the past era of cars where hardware was king, the weight of brand, scale, and quality was significant, but with AI, you will see huge changes."

He further pointed out that most of the previous automotive companies were integrated in research and development, "integrating others' capabilities to produce products that meet user needs."

Today, global software, internet, and technology companies crossing over to the automotive industry are self-developing in core areas and integrating in other areas; in addition, in the past, automakers were mainly responsible for manufacturing and then selling to dealers for sales, which made it impossible to operate, service, and charge for cars. Today, new forces in car manufacturing need to do everything from product development to after-sales service and operations Xiaopeng He believes that these two changes will lead to the final economies of scale change. Once the economies of scale of software and hardware are combined, China has the opportunity to emerge as a global automotive company. He predicts that the production and sales capacity of one million AI cars next will be the ticket to the elimination round.

"After the end-to-end model improves the lower limit capability of autonomous driving, it may only take two years to achieve super L4 standard capabilities globally. At that time, with low costs, high lower limits, and even higher upper limits, such dimensionality reduction will impact the entire autonomous driving industry," said Xiaopeng He.

However, not all automotive players have the strength to independently develop autonomous driving technology in the face of such changes. In Wu Xinzhou's view, this is an opportunity for suppliers like NVIDIA, as they can help car manufacturers fill the gaps. "We at NVIDIA are also developing large models that can be provided to car manufacturers, who can then develop on top of these models."

Wu Xinzhou stated that the presence of suppliers provides more possibilities, but it also forces all car manufacturers to undertake the same costly tasks. "Training large models and developing full-stack software are huge investments, and I believe that society as a whole may find it difficult to bear such costs."